You Only Look Yourself: Unsupervised and Untrained Single Image Dehazing Neural Network

In this paper, we study two challenging and less-touched problems in single image dehazing, namely, how to make deep learning achieve dehazing without training on the ground-truth clean (unsupervised) an collection (untrained). An unsupervised model will avoid intensive labor of collecting hazy-clean pairs, untrained is a “real” approach which could remove haze based observed hazy only no extra images are used. Motivated by layer disentanglement, propose novel method, called you look yourself (YOLY) be one first neural networks for dehazing. brief, YOLY employs three joint subnetworks separate into several latent layers, i.e., scene radiance layer, transmission map atmospheric light layer. After that, layers further composed self-supervised manner. Thanks characteristics YOLY, our method bypasses conventional paradigm models pairs or large scale dataset, thus avoids labor-intensive data domain shift issue. Besides, also provides effective learning-based transfer solution thanks its disentanglement mechanism. Extensive experiments show promising performance compared with 14 methods six databases. The code accessed at www.pengxi.me.